1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2012 Fusion-io All rights reserved. 4 * Copyright (C) 2012 Intel Corp. All rights reserved. 5 */ 6 7 #include <linux/sched.h> 8 #include <linux/bio.h> 9 #include <linux/slab.h> 10 #include <linux/blkdev.h> 11 #include <linux/raid/pq.h> 12 #include <linux/hash.h> 13 #include <linux/list_sort.h> 14 #include <linux/raid/xor.h> 15 #include <linux/mm.h> 16 #include "ctree.h" 17 #include "disk-io.h" 18 #include "volumes.h" 19 #include "raid56.h" 20 #include "async-thread.h" 21 22 /* set when additional merges to this rbio are not allowed */ 23 #define RBIO_RMW_LOCKED_BIT 1 24 25 /* 26 * set when this rbio is sitting in the hash, but it is just a cache 27 * of past RMW 28 */ 29 #define RBIO_CACHE_BIT 2 30 31 /* 32 * set when it is safe to trust the stripe_pages for caching 33 */ 34 #define RBIO_CACHE_READY_BIT 3 35 36 #define RBIO_CACHE_SIZE 1024 37 38 enum btrfs_rbio_ops { 39 BTRFS_RBIO_WRITE, 40 BTRFS_RBIO_READ_REBUILD, 41 BTRFS_RBIO_PARITY_SCRUB, 42 BTRFS_RBIO_REBUILD_MISSING, 43 }; 44 45 struct btrfs_raid_bio { 46 struct btrfs_fs_info *fs_info; 47 struct btrfs_bio *bbio; 48 49 /* while we're doing rmw on a stripe 50 * we put it into a hash table so we can 51 * lock the stripe and merge more rbios 52 * into it. 53 */ 54 struct list_head hash_list; 55 56 /* 57 * LRU list for the stripe cache 58 */ 59 struct list_head stripe_cache; 60 61 /* 62 * for scheduling work in the helper threads 63 */ 64 struct btrfs_work work; 65 66 /* 67 * bio list and bio_list_lock are used 68 * to add more bios into the stripe 69 * in hopes of avoiding the full rmw 70 */ 71 struct bio_list bio_list; 72 spinlock_t bio_list_lock; 73 74 /* also protected by the bio_list_lock, the 75 * plug list is used by the plugging code 76 * to collect partial bios while plugged. The 77 * stripe locking code also uses it to hand off 78 * the stripe lock to the next pending IO 79 */ 80 struct list_head plug_list; 81 82 /* 83 * flags that tell us if it is safe to 84 * merge with this bio 85 */ 86 unsigned long flags; 87 88 /* size of each individual stripe on disk */ 89 int stripe_len; 90 91 /* number of data stripes (no p/q) */ 92 int nr_data; 93 94 int real_stripes; 95 96 int stripe_npages; 97 /* 98 * set if we're doing a parity rebuild 99 * for a read from higher up, which is handled 100 * differently from a parity rebuild as part of 101 * rmw 102 */ 103 enum btrfs_rbio_ops operation; 104 105 /* first bad stripe */ 106 int faila; 107 108 /* second bad stripe (for raid6 use) */ 109 int failb; 110 111 int scrubp; 112 /* 113 * number of pages needed to represent the full 114 * stripe 115 */ 116 int nr_pages; 117 118 /* 119 * size of all the bios in the bio_list. This 120 * helps us decide if the rbio maps to a full 121 * stripe or not 122 */ 123 int bio_list_bytes; 124 125 int generic_bio_cnt; 126 127 refcount_t refs; 128 129 atomic_t stripes_pending; 130 131 atomic_t error; 132 /* 133 * these are two arrays of pointers. We allocate the 134 * rbio big enough to hold them both and setup their 135 * locations when the rbio is allocated 136 */ 137 138 /* pointers to pages that we allocated for 139 * reading/writing stripes directly from the disk (including P/Q) 140 */ 141 struct page **stripe_pages; 142 143 /* 144 * pointers to the pages in the bio_list. Stored 145 * here for faster lookup 146 */ 147 struct page **bio_pages; 148 149 /* 150 * bitmap to record which horizontal stripe has data 151 */ 152 unsigned long *dbitmap; 153 154 /* allocated with real_stripes-many pointers for finish_*() calls */ 155 void **finish_pointers; 156 157 /* allocated with stripe_npages-many bits for finish_*() calls */ 158 unsigned long *finish_pbitmap; 159 }; 160 161 static int __raid56_parity_recover(struct btrfs_raid_bio *rbio); 162 static noinline void finish_rmw(struct btrfs_raid_bio *rbio); 163 static void rmw_work(struct btrfs_work *work); 164 static void read_rebuild_work(struct btrfs_work *work); 165 static int fail_bio_stripe(struct btrfs_raid_bio *rbio, struct bio *bio); 166 static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed); 167 static void __free_raid_bio(struct btrfs_raid_bio *rbio); 168 static void index_rbio_pages(struct btrfs_raid_bio *rbio); 169 static int alloc_rbio_pages(struct btrfs_raid_bio *rbio); 170 171 static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, 172 int need_check); 173 static void scrub_parity_work(struct btrfs_work *work); 174 175 static void start_async_work(struct btrfs_raid_bio *rbio, btrfs_func_t work_func) 176 { 177 btrfs_init_work(&rbio->work, btrfs_rmw_helper, work_func, NULL, NULL); 178 btrfs_queue_work(rbio->fs_info->rmw_workers, &rbio->work); 179 } 180 181 /* 182 * the stripe hash table is used for locking, and to collect 183 * bios in hopes of making a full stripe 184 */ 185 int btrfs_alloc_stripe_hash_table(struct btrfs_fs_info *info) 186 { 187 struct btrfs_stripe_hash_table *table; 188 struct btrfs_stripe_hash_table *x; 189 struct btrfs_stripe_hash *cur; 190 struct btrfs_stripe_hash *h; 191 int num_entries = 1 << BTRFS_STRIPE_HASH_TABLE_BITS; 192 int i; 193 int table_size; 194 195 if (info->stripe_hash_table) 196 return 0; 197 198 /* 199 * The table is large, starting with order 4 and can go as high as 200 * order 7 in case lock debugging is turned on. 201 * 202 * Try harder to allocate and fallback to vmalloc to lower the chance 203 * of a failing mount. 204 */ 205 table_size = sizeof(*table) + sizeof(*h) * num_entries; 206 table = kvzalloc(table_size, GFP_KERNEL); 207 if (!table) 208 return -ENOMEM; 209 210 spin_lock_init(&table->cache_lock); 211 INIT_LIST_HEAD(&table->stripe_cache); 212 213 h = table->table; 214 215 for (i = 0; i < num_entries; i++) { 216 cur = h + i; 217 INIT_LIST_HEAD(&cur->hash_list); 218 spin_lock_init(&cur->lock); 219 } 220 221 x = cmpxchg(&info->stripe_hash_table, NULL, table); 222 if (x) 223 kvfree(x); 224 return 0; 225 } 226 227 /* 228 * caching an rbio means to copy anything from the 229 * bio_pages array into the stripe_pages array. We 230 * use the page uptodate bit in the stripe cache array 231 * to indicate if it has valid data 232 * 233 * once the caching is done, we set the cache ready 234 * bit. 235 */ 236 static void cache_rbio_pages(struct btrfs_raid_bio *rbio) 237 { 238 int i; 239 char *s; 240 char *d; 241 int ret; 242 243 ret = alloc_rbio_pages(rbio); 244 if (ret) 245 return; 246 247 for (i = 0; i < rbio->nr_pages; i++) { 248 if (!rbio->bio_pages[i]) 249 continue; 250 251 s = kmap(rbio->bio_pages[i]); 252 d = kmap(rbio->stripe_pages[i]); 253 254 copy_page(d, s); 255 256 kunmap(rbio->bio_pages[i]); 257 kunmap(rbio->stripe_pages[i]); 258 SetPageUptodate(rbio->stripe_pages[i]); 259 } 260 set_bit(RBIO_CACHE_READY_BIT, &rbio->flags); 261 } 262 263 /* 264 * we hash on the first logical address of the stripe 265 */ 266 static int rbio_bucket(struct btrfs_raid_bio *rbio) 267 { 268 u64 num = rbio->bbio->raid_map[0]; 269 270 /* 271 * we shift down quite a bit. We're using byte 272 * addressing, and most of the lower bits are zeros. 273 * This tends to upset hash_64, and it consistently 274 * returns just one or two different values. 275 * 276 * shifting off the lower bits fixes things. 277 */ 278 return hash_64(num >> 16, BTRFS_STRIPE_HASH_TABLE_BITS); 279 } 280 281 /* 282 * stealing an rbio means taking all the uptodate pages from the stripe 283 * array in the source rbio and putting them into the destination rbio 284 */ 285 static void steal_rbio(struct btrfs_raid_bio *src, struct btrfs_raid_bio *dest) 286 { 287 int i; 288 struct page *s; 289 struct page *d; 290 291 if (!test_bit(RBIO_CACHE_READY_BIT, &src->flags)) 292 return; 293 294 for (i = 0; i < dest->nr_pages; i++) { 295 s = src->stripe_pages[i]; 296 if (!s || !PageUptodate(s)) { 297 continue; 298 } 299 300 d = dest->stripe_pages[i]; 301 if (d) 302 __free_page(d); 303 304 dest->stripe_pages[i] = s; 305 src->stripe_pages[i] = NULL; 306 } 307 } 308 309 /* 310 * merging means we take the bio_list from the victim and 311 * splice it into the destination. The victim should 312 * be discarded afterwards. 313 * 314 * must be called with dest->rbio_list_lock held 315 */ 316 static void merge_rbio(struct btrfs_raid_bio *dest, 317 struct btrfs_raid_bio *victim) 318 { 319 bio_list_merge(&dest->bio_list, &victim->bio_list); 320 dest->bio_list_bytes += victim->bio_list_bytes; 321 dest->generic_bio_cnt += victim->generic_bio_cnt; 322 bio_list_init(&victim->bio_list); 323 } 324 325 /* 326 * used to prune items that are in the cache. The caller 327 * must hold the hash table lock. 328 */ 329 static void __remove_rbio_from_cache(struct btrfs_raid_bio *rbio) 330 { 331 int bucket = rbio_bucket(rbio); 332 struct btrfs_stripe_hash_table *table; 333 struct btrfs_stripe_hash *h; 334 int freeit = 0; 335 336 /* 337 * check the bit again under the hash table lock. 338 */ 339 if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) 340 return; 341 342 table = rbio->fs_info->stripe_hash_table; 343 h = table->table + bucket; 344 345 /* hold the lock for the bucket because we may be 346 * removing it from the hash table 347 */ 348 spin_lock(&h->lock); 349 350 /* 351 * hold the lock for the bio list because we need 352 * to make sure the bio list is empty 353 */ 354 spin_lock(&rbio->bio_list_lock); 355 356 if (test_and_clear_bit(RBIO_CACHE_BIT, &rbio->flags)) { 357 list_del_init(&rbio->stripe_cache); 358 table->cache_size -= 1; 359 freeit = 1; 360 361 /* if the bio list isn't empty, this rbio is 362 * still involved in an IO. We take it out 363 * of the cache list, and drop the ref that 364 * was held for the list. 365 * 366 * If the bio_list was empty, we also remove 367 * the rbio from the hash_table, and drop 368 * the corresponding ref 369 */ 370 if (bio_list_empty(&rbio->bio_list)) { 371 if (!list_empty(&rbio->hash_list)) { 372 list_del_init(&rbio->hash_list); 373 refcount_dec(&rbio->refs); 374 BUG_ON(!list_empty(&rbio->plug_list)); 375 } 376 } 377 } 378 379 spin_unlock(&rbio->bio_list_lock); 380 spin_unlock(&h->lock); 381 382 if (freeit) 383 __free_raid_bio(rbio); 384 } 385 386 /* 387 * prune a given rbio from the cache 388 */ 389 static void remove_rbio_from_cache(struct btrfs_raid_bio *rbio) 390 { 391 struct btrfs_stripe_hash_table *table; 392 unsigned long flags; 393 394 if (!test_bit(RBIO_CACHE_BIT, &rbio->flags)) 395 return; 396 397 table = rbio->fs_info->stripe_hash_table; 398 399 spin_lock_irqsave(&table->cache_lock, flags); 400 __remove_rbio_from_cache(rbio); 401 spin_unlock_irqrestore(&table->cache_lock, flags); 402 } 403 404 /* 405 * remove everything in the cache 406 */ 407 static void btrfs_clear_rbio_cache(struct btrfs_fs_info *info) 408 { 409 struct btrfs_stripe_hash_table *table; 410 unsigned long flags; 411 struct btrfs_raid_bio *rbio; 412 413 table = info->stripe_hash_table; 414 415 spin_lock_irqsave(&table->cache_lock, flags); 416 while (!list_empty(&table->stripe_cache)) { 417 rbio = list_entry(table->stripe_cache.next, 418 struct btrfs_raid_bio, 419 stripe_cache); 420 __remove_rbio_from_cache(rbio); 421 } 422 spin_unlock_irqrestore(&table->cache_lock, flags); 423 } 424 425 /* 426 * remove all cached entries and free the hash table 427 * used by unmount 428 */ 429 void btrfs_free_stripe_hash_table(struct btrfs_fs_info *info) 430 { 431 if (!info->stripe_hash_table) 432 return; 433 btrfs_clear_rbio_cache(info); 434 kvfree(info->stripe_hash_table); 435 info->stripe_hash_table = NULL; 436 } 437 438 /* 439 * insert an rbio into the stripe cache. It 440 * must have already been prepared by calling 441 * cache_rbio_pages 442 * 443 * If this rbio was already cached, it gets 444 * moved to the front of the lru. 445 * 446 * If the size of the rbio cache is too big, we 447 * prune an item. 448 */ 449 static void cache_rbio(struct btrfs_raid_bio *rbio) 450 { 451 struct btrfs_stripe_hash_table *table; 452 unsigned long flags; 453 454 if (!test_bit(RBIO_CACHE_READY_BIT, &rbio->flags)) 455 return; 456 457 table = rbio->fs_info->stripe_hash_table; 458 459 spin_lock_irqsave(&table->cache_lock, flags); 460 spin_lock(&rbio->bio_list_lock); 461 462 /* bump our ref if we were not in the list before */ 463 if (!test_and_set_bit(RBIO_CACHE_BIT, &rbio->flags)) 464 refcount_inc(&rbio->refs); 465 466 if (!list_empty(&rbio->stripe_cache)){ 467 list_move(&rbio->stripe_cache, &table->stripe_cache); 468 } else { 469 list_add(&rbio->stripe_cache, &table->stripe_cache); 470 table->cache_size += 1; 471 } 472 473 spin_unlock(&rbio->bio_list_lock); 474 475 if (table->cache_size > RBIO_CACHE_SIZE) { 476 struct btrfs_raid_bio *found; 477 478 found = list_entry(table->stripe_cache.prev, 479 struct btrfs_raid_bio, 480 stripe_cache); 481 482 if (found != rbio) 483 __remove_rbio_from_cache(found); 484 } 485 486 spin_unlock_irqrestore(&table->cache_lock, flags); 487 } 488 489 /* 490 * helper function to run the xor_blocks api. It is only 491 * able to do MAX_XOR_BLOCKS at a time, so we need to 492 * loop through. 493 */ 494 static void run_xor(void **pages, int src_cnt, ssize_t len) 495 { 496 int src_off = 0; 497 int xor_src_cnt = 0; 498 void *dest = pages[src_cnt]; 499 500 while(src_cnt > 0) { 501 xor_src_cnt = min(src_cnt, MAX_XOR_BLOCKS); 502 xor_blocks(xor_src_cnt, len, dest, pages + src_off); 503 504 src_cnt -= xor_src_cnt; 505 src_off += xor_src_cnt; 506 } 507 } 508 509 /* 510 * Returns true if the bio list inside this rbio covers an entire stripe (no 511 * rmw required). 512 */ 513 static int rbio_is_full(struct btrfs_raid_bio *rbio) 514 { 515 unsigned long flags; 516 unsigned long size = rbio->bio_list_bytes; 517 int ret = 1; 518 519 spin_lock_irqsave(&rbio->bio_list_lock, flags); 520 if (size != rbio->nr_data * rbio->stripe_len) 521 ret = 0; 522 BUG_ON(size > rbio->nr_data * rbio->stripe_len); 523 spin_unlock_irqrestore(&rbio->bio_list_lock, flags); 524 525 return ret; 526 } 527 528 /* 529 * returns 1 if it is safe to merge two rbios together. 530 * The merging is safe if the two rbios correspond to 531 * the same stripe and if they are both going in the same 532 * direction (read vs write), and if neither one is 533 * locked for final IO 534 * 535 * The caller is responsible for locking such that 536 * rmw_locked is safe to test 537 */ 538 static int rbio_can_merge(struct btrfs_raid_bio *last, 539 struct btrfs_raid_bio *cur) 540 { 541 if (test_bit(RBIO_RMW_LOCKED_BIT, &last->flags) || 542 test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) 543 return 0; 544 545 /* 546 * we can't merge with cached rbios, since the 547 * idea is that when we merge the destination 548 * rbio is going to run our IO for us. We can 549 * steal from cached rbios though, other functions 550 * handle that. 551 */ 552 if (test_bit(RBIO_CACHE_BIT, &last->flags) || 553 test_bit(RBIO_CACHE_BIT, &cur->flags)) 554 return 0; 555 556 if (last->bbio->raid_map[0] != 557 cur->bbio->raid_map[0]) 558 return 0; 559 560 /* we can't merge with different operations */ 561 if (last->operation != cur->operation) 562 return 0; 563 /* 564 * We've need read the full stripe from the drive. 565 * check and repair the parity and write the new results. 566 * 567 * We're not allowed to add any new bios to the 568 * bio list here, anyone else that wants to 569 * change this stripe needs to do their own rmw. 570 */ 571 if (last->operation == BTRFS_RBIO_PARITY_SCRUB) 572 return 0; 573 574 if (last->operation == BTRFS_RBIO_REBUILD_MISSING) 575 return 0; 576 577 if (last->operation == BTRFS_RBIO_READ_REBUILD) { 578 int fa = last->faila; 579 int fb = last->failb; 580 int cur_fa = cur->faila; 581 int cur_fb = cur->failb; 582 583 if (last->faila >= last->failb) { 584 fa = last->failb; 585 fb = last->faila; 586 } 587 588 if (cur->faila >= cur->failb) { 589 cur_fa = cur->failb; 590 cur_fb = cur->faila; 591 } 592 593 if (fa != cur_fa || fb != cur_fb) 594 return 0; 595 } 596 return 1; 597 } 598 599 static int rbio_stripe_page_index(struct btrfs_raid_bio *rbio, int stripe, 600 int index) 601 { 602 return stripe * rbio->stripe_npages + index; 603 } 604 605 /* 606 * these are just the pages from the rbio array, not from anything 607 * the FS sent down to us 608 */ 609 static struct page *rbio_stripe_page(struct btrfs_raid_bio *rbio, int stripe, 610 int index) 611 { 612 return rbio->stripe_pages[rbio_stripe_page_index(rbio, stripe, index)]; 613 } 614 615 /* 616 * helper to index into the pstripe 617 */ 618 static struct page *rbio_pstripe_page(struct btrfs_raid_bio *rbio, int index) 619 { 620 return rbio_stripe_page(rbio, rbio->nr_data, index); 621 } 622 623 /* 624 * helper to index into the qstripe, returns null 625 * if there is no qstripe 626 */ 627 static struct page *rbio_qstripe_page(struct btrfs_raid_bio *rbio, int index) 628 { 629 if (rbio->nr_data + 1 == rbio->real_stripes) 630 return NULL; 631 return rbio_stripe_page(rbio, rbio->nr_data + 1, index); 632 } 633 634 /* 635 * The first stripe in the table for a logical address 636 * has the lock. rbios are added in one of three ways: 637 * 638 * 1) Nobody has the stripe locked yet. The rbio is given 639 * the lock and 0 is returned. The caller must start the IO 640 * themselves. 641 * 642 * 2) Someone has the stripe locked, but we're able to merge 643 * with the lock owner. The rbio is freed and the IO will 644 * start automatically along with the existing rbio. 1 is returned. 645 * 646 * 3) Someone has the stripe locked, but we're not able to merge. 647 * The rbio is added to the lock owner's plug list, or merged into 648 * an rbio already on the plug list. When the lock owner unlocks, 649 * the next rbio on the list is run and the IO is started automatically. 650 * 1 is returned 651 * 652 * If we return 0, the caller still owns the rbio and must continue with 653 * IO submission. If we return 1, the caller must assume the rbio has 654 * already been freed. 655 */ 656 static noinline int lock_stripe_add(struct btrfs_raid_bio *rbio) 657 { 658 int bucket = rbio_bucket(rbio); 659 struct btrfs_stripe_hash *h = rbio->fs_info->stripe_hash_table->table + bucket; 660 struct btrfs_raid_bio *cur; 661 struct btrfs_raid_bio *pending; 662 unsigned long flags; 663 struct btrfs_raid_bio *freeit = NULL; 664 struct btrfs_raid_bio *cache_drop = NULL; 665 int ret = 0; 666 667 spin_lock_irqsave(&h->lock, flags); 668 list_for_each_entry(cur, &h->hash_list, hash_list) { 669 if (cur->bbio->raid_map[0] == rbio->bbio->raid_map[0]) { 670 spin_lock(&cur->bio_list_lock); 671 672 /* can we steal this cached rbio's pages? */ 673 if (bio_list_empty(&cur->bio_list) && 674 list_empty(&cur->plug_list) && 675 test_bit(RBIO_CACHE_BIT, &cur->flags) && 676 !test_bit(RBIO_RMW_LOCKED_BIT, &cur->flags)) { 677 list_del_init(&cur->hash_list); 678 refcount_dec(&cur->refs); 679 680 steal_rbio(cur, rbio); 681 cache_drop = cur; 682 spin_unlock(&cur->bio_list_lock); 683 684 goto lockit; 685 } 686 687 /* can we merge into the lock owner? */ 688 if (rbio_can_merge(cur, rbio)) { 689 merge_rbio(cur, rbio); 690 spin_unlock(&cur->bio_list_lock); 691 freeit = rbio; 692 ret = 1; 693 goto out; 694 } 695 696 697 /* 698 * we couldn't merge with the running 699 * rbio, see if we can merge with the 700 * pending ones. We don't have to 701 * check for rmw_locked because there 702 * is no way they are inside finish_rmw 703 * right now 704 */ 705 list_for_each_entry(pending, &cur->plug_list, 706 plug_list) { 707 if (rbio_can_merge(pending, rbio)) { 708 merge_rbio(pending, rbio); 709 spin_unlock(&cur->bio_list_lock); 710 freeit = rbio; 711 ret = 1; 712 goto out; 713 } 714 } 715 716 /* no merging, put us on the tail of the plug list, 717 * our rbio will be started with the currently 718 * running rbio unlocks 719 */ 720 list_add_tail(&rbio->plug_list, &cur->plug_list); 721 spin_unlock(&cur->bio_list_lock); 722 ret = 1; 723 goto out; 724 } 725 } 726 lockit: 727 refcount_inc(&rbio->refs); 728 list_add(&rbio->hash_list, &h->hash_list); 729 out: 730 spin_unlock_irqrestore(&h->lock, flags); 731 if (cache_drop) 732 remove_rbio_from_cache(cache_drop); 733 if (freeit) 734 __free_raid_bio(freeit); 735 return ret; 736 } 737 738 /* 739 * called as rmw or parity rebuild is completed. If the plug list has more 740 * rbios waiting for this stripe, the next one on the list will be started 741 */ 742 static noinline void unlock_stripe(struct btrfs_raid_bio *rbio) 743 { 744 int bucket; 745 struct btrfs_stripe_hash *h; 746 unsigned long flags; 747 int keep_cache = 0; 748 749 bucket = rbio_bucket(rbio); 750 h = rbio->fs_info->stripe_hash_table->table + bucket; 751 752 if (list_empty(&rbio->plug_list)) 753 cache_rbio(rbio); 754 755 spin_lock_irqsave(&h->lock, flags); 756 spin_lock(&rbio->bio_list_lock); 757 758 if (!list_empty(&rbio->hash_list)) { 759 /* 760 * if we're still cached and there is no other IO 761 * to perform, just leave this rbio here for others 762 * to steal from later 763 */ 764 if (list_empty(&rbio->plug_list) && 765 test_bit(RBIO_CACHE_BIT, &rbio->flags)) { 766 keep_cache = 1; 767 clear_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); 768 BUG_ON(!bio_list_empty(&rbio->bio_list)); 769 goto done; 770 } 771 772 list_del_init(&rbio->hash_list); 773 refcount_dec(&rbio->refs); 774 775 /* 776 * we use the plug list to hold all the rbios 777 * waiting for the chance to lock this stripe. 778 * hand the lock over to one of them. 779 */ 780 if (!list_empty(&rbio->plug_list)) { 781 struct btrfs_raid_bio *next; 782 struct list_head *head = rbio->plug_list.next; 783 784 next = list_entry(head, struct btrfs_raid_bio, 785 plug_list); 786 787 list_del_init(&rbio->plug_list); 788 789 list_add(&next->hash_list, &h->hash_list); 790 refcount_inc(&next->refs); 791 spin_unlock(&rbio->bio_list_lock); 792 spin_unlock_irqrestore(&h->lock, flags); 793 794 if (next->operation == BTRFS_RBIO_READ_REBUILD) 795 start_async_work(next, read_rebuild_work); 796 else if (next->operation == BTRFS_RBIO_REBUILD_MISSING) { 797 steal_rbio(rbio, next); 798 start_async_work(next, read_rebuild_work); 799 } else if (next->operation == BTRFS_RBIO_WRITE) { 800 steal_rbio(rbio, next); 801 start_async_work(next, rmw_work); 802 } else if (next->operation == BTRFS_RBIO_PARITY_SCRUB) { 803 steal_rbio(rbio, next); 804 start_async_work(next, scrub_parity_work); 805 } 806 807 goto done_nolock; 808 } 809 } 810 done: 811 spin_unlock(&rbio->bio_list_lock); 812 spin_unlock_irqrestore(&h->lock, flags); 813 814 done_nolock: 815 if (!keep_cache) 816 remove_rbio_from_cache(rbio); 817 } 818 819 static void __free_raid_bio(struct btrfs_raid_bio *rbio) 820 { 821 int i; 822 823 if (!refcount_dec_and_test(&rbio->refs)) 824 return; 825 826 WARN_ON(!list_empty(&rbio->stripe_cache)); 827 WARN_ON(!list_empty(&rbio->hash_list)); 828 WARN_ON(!bio_list_empty(&rbio->bio_list)); 829 830 for (i = 0; i < rbio->nr_pages; i++) { 831 if (rbio->stripe_pages[i]) { 832 __free_page(rbio->stripe_pages[i]); 833 rbio->stripe_pages[i] = NULL; 834 } 835 } 836 837 btrfs_put_bbio(rbio->bbio); 838 kfree(rbio); 839 } 840 841 static void rbio_endio_bio_list(struct bio *cur, blk_status_t err) 842 { 843 struct bio *next; 844 845 while (cur) { 846 next = cur->bi_next; 847 cur->bi_next = NULL; 848 cur->bi_status = err; 849 bio_endio(cur); 850 cur = next; 851 } 852 } 853 854 /* 855 * this frees the rbio and runs through all the bios in the 856 * bio_list and calls end_io on them 857 */ 858 static void rbio_orig_end_io(struct btrfs_raid_bio *rbio, blk_status_t err) 859 { 860 struct bio *cur = bio_list_get(&rbio->bio_list); 861 struct bio *extra; 862 863 if (rbio->generic_bio_cnt) 864 btrfs_bio_counter_sub(rbio->fs_info, rbio->generic_bio_cnt); 865 866 /* 867 * At this moment, rbio->bio_list is empty, however since rbio does not 868 * always have RBIO_RMW_LOCKED_BIT set and rbio is still linked on the 869 * hash list, rbio may be merged with others so that rbio->bio_list 870 * becomes non-empty. 871 * Once unlock_stripe() is done, rbio->bio_list will not be updated any 872 * more and we can call bio_endio() on all queued bios. 873 */ 874 unlock_stripe(rbio); 875 extra = bio_list_get(&rbio->bio_list); 876 __free_raid_bio(rbio); 877 878 rbio_endio_bio_list(cur, err); 879 if (extra) 880 rbio_endio_bio_list(extra, err); 881 } 882 883 /* 884 * end io function used by finish_rmw. When we finally 885 * get here, we've written a full stripe 886 */ 887 static void raid_write_end_io(struct bio *bio) 888 { 889 struct btrfs_raid_bio *rbio = bio->bi_private; 890 blk_status_t err = bio->bi_status; 891 int max_errors; 892 893 if (err) 894 fail_bio_stripe(rbio, bio); 895 896 bio_put(bio); 897 898 if (!atomic_dec_and_test(&rbio->stripes_pending)) 899 return; 900 901 err = BLK_STS_OK; 902 903 /* OK, we have read all the stripes we need to. */ 904 max_errors = (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) ? 905 0 : rbio->bbio->max_errors; 906 if (atomic_read(&rbio->error) > max_errors) 907 err = BLK_STS_IOERR; 908 909 rbio_orig_end_io(rbio, err); 910 } 911 912 /* 913 * the read/modify/write code wants to use the original bio for 914 * any pages it included, and then use the rbio for everything 915 * else. This function decides if a given index (stripe number) 916 * and page number in that stripe fall inside the original bio 917 * or the rbio. 918 * 919 * if you set bio_list_only, you'll get a NULL back for any ranges 920 * that are outside the bio_list 921 * 922 * This doesn't take any refs on anything, you get a bare page pointer 923 * and the caller must bump refs as required. 924 * 925 * You must call index_rbio_pages once before you can trust 926 * the answers from this function. 927 */ 928 static struct page *page_in_rbio(struct btrfs_raid_bio *rbio, 929 int index, int pagenr, int bio_list_only) 930 { 931 int chunk_page; 932 struct page *p = NULL; 933 934 chunk_page = index * (rbio->stripe_len >> PAGE_SHIFT) + pagenr; 935 936 spin_lock_irq(&rbio->bio_list_lock); 937 p = rbio->bio_pages[chunk_page]; 938 spin_unlock_irq(&rbio->bio_list_lock); 939 940 if (p || bio_list_only) 941 return p; 942 943 return rbio->stripe_pages[chunk_page]; 944 } 945 946 /* 947 * number of pages we need for the entire stripe across all the 948 * drives 949 */ 950 static unsigned long rbio_nr_pages(unsigned long stripe_len, int nr_stripes) 951 { 952 return DIV_ROUND_UP(stripe_len, PAGE_SIZE) * nr_stripes; 953 } 954 955 /* 956 * allocation and initial setup for the btrfs_raid_bio. Not 957 * this does not allocate any pages for rbio->pages. 958 */ 959 static struct btrfs_raid_bio *alloc_rbio(struct btrfs_fs_info *fs_info, 960 struct btrfs_bio *bbio, 961 u64 stripe_len) 962 { 963 struct btrfs_raid_bio *rbio; 964 int nr_data = 0; 965 int real_stripes = bbio->num_stripes - bbio->num_tgtdevs; 966 int num_pages = rbio_nr_pages(stripe_len, real_stripes); 967 int stripe_npages = DIV_ROUND_UP(stripe_len, PAGE_SIZE); 968 void *p; 969 970 rbio = kzalloc(sizeof(*rbio) + 971 sizeof(*rbio->stripe_pages) * num_pages + 972 sizeof(*rbio->bio_pages) * num_pages + 973 sizeof(*rbio->finish_pointers) * real_stripes + 974 sizeof(*rbio->dbitmap) * BITS_TO_LONGS(stripe_npages) + 975 sizeof(*rbio->finish_pbitmap) * 976 BITS_TO_LONGS(stripe_npages), 977 GFP_NOFS); 978 if (!rbio) 979 return ERR_PTR(-ENOMEM); 980 981 bio_list_init(&rbio->bio_list); 982 INIT_LIST_HEAD(&rbio->plug_list); 983 spin_lock_init(&rbio->bio_list_lock); 984 INIT_LIST_HEAD(&rbio->stripe_cache); 985 INIT_LIST_HEAD(&rbio->hash_list); 986 rbio->bbio = bbio; 987 rbio->fs_info = fs_info; 988 rbio->stripe_len = stripe_len; 989 rbio->nr_pages = num_pages; 990 rbio->real_stripes = real_stripes; 991 rbio->stripe_npages = stripe_npages; 992 rbio->faila = -1; 993 rbio->failb = -1; 994 refcount_set(&rbio->refs, 1); 995 atomic_set(&rbio->error, 0); 996 atomic_set(&rbio->stripes_pending, 0); 997 998 /* 999 * the stripe_pages, bio_pages, etc arrays point to the extra 1000 * memory we allocated past the end of the rbio 1001 */ 1002 p = rbio + 1; 1003 #define CONSUME_ALLOC(ptr, count) do { \ 1004 ptr = p; \ 1005 p = (unsigned char *)p + sizeof(*(ptr)) * (count); \ 1006 } while (0) 1007 CONSUME_ALLOC(rbio->stripe_pages, num_pages); 1008 CONSUME_ALLOC(rbio->bio_pages, num_pages); 1009 CONSUME_ALLOC(rbio->finish_pointers, real_stripes); 1010 CONSUME_ALLOC(rbio->dbitmap, BITS_TO_LONGS(stripe_npages)); 1011 CONSUME_ALLOC(rbio->finish_pbitmap, BITS_TO_LONGS(stripe_npages)); 1012 #undef CONSUME_ALLOC 1013 1014 if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID5) 1015 nr_data = real_stripes - 1; 1016 else if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) 1017 nr_data = real_stripes - 2; 1018 else 1019 BUG(); 1020 1021 rbio->nr_data = nr_data; 1022 return rbio; 1023 } 1024 1025 /* allocate pages for all the stripes in the bio, including parity */ 1026 static int alloc_rbio_pages(struct btrfs_raid_bio *rbio) 1027 { 1028 int i; 1029 struct page *page; 1030 1031 for (i = 0; i < rbio->nr_pages; i++) { 1032 if (rbio->stripe_pages[i]) 1033 continue; 1034 page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 1035 if (!page) 1036 return -ENOMEM; 1037 rbio->stripe_pages[i] = page; 1038 } 1039 return 0; 1040 } 1041 1042 /* only allocate pages for p/q stripes */ 1043 static int alloc_rbio_parity_pages(struct btrfs_raid_bio *rbio) 1044 { 1045 int i; 1046 struct page *page; 1047 1048 i = rbio_stripe_page_index(rbio, rbio->nr_data, 0); 1049 1050 for (; i < rbio->nr_pages; i++) { 1051 if (rbio->stripe_pages[i]) 1052 continue; 1053 page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 1054 if (!page) 1055 return -ENOMEM; 1056 rbio->stripe_pages[i] = page; 1057 } 1058 return 0; 1059 } 1060 1061 /* 1062 * add a single page from a specific stripe into our list of bios for IO 1063 * this will try to merge into existing bios if possible, and returns 1064 * zero if all went well. 1065 */ 1066 static int rbio_add_io_page(struct btrfs_raid_bio *rbio, 1067 struct bio_list *bio_list, 1068 struct page *page, 1069 int stripe_nr, 1070 unsigned long page_index, 1071 unsigned long bio_max_len) 1072 { 1073 struct bio *last = bio_list->tail; 1074 u64 last_end = 0; 1075 int ret; 1076 struct bio *bio; 1077 struct btrfs_bio_stripe *stripe; 1078 u64 disk_start; 1079 1080 stripe = &rbio->bbio->stripes[stripe_nr]; 1081 disk_start = stripe->physical + (page_index << PAGE_SHIFT); 1082 1083 /* if the device is missing, just fail this stripe */ 1084 if (!stripe->dev->bdev) 1085 return fail_rbio_index(rbio, stripe_nr); 1086 1087 /* see if we can add this page onto our existing bio */ 1088 if (last) { 1089 last_end = (u64)last->bi_iter.bi_sector << 9; 1090 last_end += last->bi_iter.bi_size; 1091 1092 /* 1093 * we can't merge these if they are from different 1094 * devices or if they are not contiguous 1095 */ 1096 if (last_end == disk_start && stripe->dev->bdev && 1097 !last->bi_status && 1098 last->bi_disk == stripe->dev->bdev->bd_disk && 1099 last->bi_partno == stripe->dev->bdev->bd_partno) { 1100 ret = bio_add_page(last, page, PAGE_SIZE, 0); 1101 if (ret == PAGE_SIZE) 1102 return 0; 1103 } 1104 } 1105 1106 /* put a new bio on the list */ 1107 bio = btrfs_io_bio_alloc(bio_max_len >> PAGE_SHIFT ?: 1); 1108 bio->bi_iter.bi_size = 0; 1109 bio_set_dev(bio, stripe->dev->bdev); 1110 bio->bi_iter.bi_sector = disk_start >> 9; 1111 1112 bio_add_page(bio, page, PAGE_SIZE, 0); 1113 bio_list_add(bio_list, bio); 1114 return 0; 1115 } 1116 1117 /* 1118 * while we're doing the read/modify/write cycle, we could 1119 * have errors in reading pages off the disk. This checks 1120 * for errors and if we're not able to read the page it'll 1121 * trigger parity reconstruction. The rmw will be finished 1122 * after we've reconstructed the failed stripes 1123 */ 1124 static void validate_rbio_for_rmw(struct btrfs_raid_bio *rbio) 1125 { 1126 if (rbio->faila >= 0 || rbio->failb >= 0) { 1127 BUG_ON(rbio->faila == rbio->real_stripes - 1); 1128 __raid56_parity_recover(rbio); 1129 } else { 1130 finish_rmw(rbio); 1131 } 1132 } 1133 1134 /* 1135 * helper function to walk our bio list and populate the bio_pages array with 1136 * the result. This seems expensive, but it is faster than constantly 1137 * searching through the bio list as we setup the IO in finish_rmw or stripe 1138 * reconstruction. 1139 * 1140 * This must be called before you trust the answers from page_in_rbio 1141 */ 1142 static void index_rbio_pages(struct btrfs_raid_bio *rbio) 1143 { 1144 struct bio *bio; 1145 u64 start; 1146 unsigned long stripe_offset; 1147 unsigned long page_index; 1148 1149 spin_lock_irq(&rbio->bio_list_lock); 1150 bio_list_for_each(bio, &rbio->bio_list) { 1151 struct bio_vec bvec; 1152 struct bvec_iter iter; 1153 int i = 0; 1154 1155 start = (u64)bio->bi_iter.bi_sector << 9; 1156 stripe_offset = start - rbio->bbio->raid_map[0]; 1157 page_index = stripe_offset >> PAGE_SHIFT; 1158 1159 if (bio_flagged(bio, BIO_CLONED)) 1160 bio->bi_iter = btrfs_io_bio(bio)->iter; 1161 1162 bio_for_each_segment(bvec, bio, iter) { 1163 rbio->bio_pages[page_index + i] = bvec.bv_page; 1164 i++; 1165 } 1166 } 1167 spin_unlock_irq(&rbio->bio_list_lock); 1168 } 1169 1170 /* 1171 * this is called from one of two situations. We either 1172 * have a full stripe from the higher layers, or we've read all 1173 * the missing bits off disk. 1174 * 1175 * This will calculate the parity and then send down any 1176 * changed blocks. 1177 */ 1178 static noinline void finish_rmw(struct btrfs_raid_bio *rbio) 1179 { 1180 struct btrfs_bio *bbio = rbio->bbio; 1181 void **pointers = rbio->finish_pointers; 1182 int nr_data = rbio->nr_data; 1183 int stripe; 1184 int pagenr; 1185 int p_stripe = -1; 1186 int q_stripe = -1; 1187 struct bio_list bio_list; 1188 struct bio *bio; 1189 int ret; 1190 1191 bio_list_init(&bio_list); 1192 1193 if (rbio->real_stripes - rbio->nr_data == 1) { 1194 p_stripe = rbio->real_stripes - 1; 1195 } else if (rbio->real_stripes - rbio->nr_data == 2) { 1196 p_stripe = rbio->real_stripes - 2; 1197 q_stripe = rbio->real_stripes - 1; 1198 } else { 1199 BUG(); 1200 } 1201 1202 /* at this point we either have a full stripe, 1203 * or we've read the full stripe from the drive. 1204 * recalculate the parity and write the new results. 1205 * 1206 * We're not allowed to add any new bios to the 1207 * bio list here, anyone else that wants to 1208 * change this stripe needs to do their own rmw. 1209 */ 1210 spin_lock_irq(&rbio->bio_list_lock); 1211 set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); 1212 spin_unlock_irq(&rbio->bio_list_lock); 1213 1214 atomic_set(&rbio->error, 0); 1215 1216 /* 1217 * now that we've set rmw_locked, run through the 1218 * bio list one last time and map the page pointers 1219 * 1220 * We don't cache full rbios because we're assuming 1221 * the higher layers are unlikely to use this area of 1222 * the disk again soon. If they do use it again, 1223 * hopefully they will send another full bio. 1224 */ 1225 index_rbio_pages(rbio); 1226 if (!rbio_is_full(rbio)) 1227 cache_rbio_pages(rbio); 1228 else 1229 clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); 1230 1231 for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { 1232 struct page *p; 1233 /* first collect one page from each data stripe */ 1234 for (stripe = 0; stripe < nr_data; stripe++) { 1235 p = page_in_rbio(rbio, stripe, pagenr, 0); 1236 pointers[stripe] = kmap(p); 1237 } 1238 1239 /* then add the parity stripe */ 1240 p = rbio_pstripe_page(rbio, pagenr); 1241 SetPageUptodate(p); 1242 pointers[stripe++] = kmap(p); 1243 1244 if (q_stripe != -1) { 1245 1246 /* 1247 * raid6, add the qstripe and call the 1248 * library function to fill in our p/q 1249 */ 1250 p = rbio_qstripe_page(rbio, pagenr); 1251 SetPageUptodate(p); 1252 pointers[stripe++] = kmap(p); 1253 1254 raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, 1255 pointers); 1256 } else { 1257 /* raid5 */ 1258 copy_page(pointers[nr_data], pointers[0]); 1259 run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); 1260 } 1261 1262 1263 for (stripe = 0; stripe < rbio->real_stripes; stripe++) 1264 kunmap(page_in_rbio(rbio, stripe, pagenr, 0)); 1265 } 1266 1267 /* 1268 * time to start writing. Make bios for everything from the 1269 * higher layers (the bio_list in our rbio) and our p/q. Ignore 1270 * everything else. 1271 */ 1272 for (stripe = 0; stripe < rbio->real_stripes; stripe++) { 1273 for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { 1274 struct page *page; 1275 if (stripe < rbio->nr_data) { 1276 page = page_in_rbio(rbio, stripe, pagenr, 1); 1277 if (!page) 1278 continue; 1279 } else { 1280 page = rbio_stripe_page(rbio, stripe, pagenr); 1281 } 1282 1283 ret = rbio_add_io_page(rbio, &bio_list, 1284 page, stripe, pagenr, rbio->stripe_len); 1285 if (ret) 1286 goto cleanup; 1287 } 1288 } 1289 1290 if (likely(!bbio->num_tgtdevs)) 1291 goto write_data; 1292 1293 for (stripe = 0; stripe < rbio->real_stripes; stripe++) { 1294 if (!bbio->tgtdev_map[stripe]) 1295 continue; 1296 1297 for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { 1298 struct page *page; 1299 if (stripe < rbio->nr_data) { 1300 page = page_in_rbio(rbio, stripe, pagenr, 1); 1301 if (!page) 1302 continue; 1303 } else { 1304 page = rbio_stripe_page(rbio, stripe, pagenr); 1305 } 1306 1307 ret = rbio_add_io_page(rbio, &bio_list, page, 1308 rbio->bbio->tgtdev_map[stripe], 1309 pagenr, rbio->stripe_len); 1310 if (ret) 1311 goto cleanup; 1312 } 1313 } 1314 1315 write_data: 1316 atomic_set(&rbio->stripes_pending, bio_list_size(&bio_list)); 1317 BUG_ON(atomic_read(&rbio->stripes_pending) == 0); 1318 1319 while (1) { 1320 bio = bio_list_pop(&bio_list); 1321 if (!bio) 1322 break; 1323 1324 bio->bi_private = rbio; 1325 bio->bi_end_io = raid_write_end_io; 1326 bio->bi_opf = REQ_OP_WRITE; 1327 1328 submit_bio(bio); 1329 } 1330 return; 1331 1332 cleanup: 1333 rbio_orig_end_io(rbio, BLK_STS_IOERR); 1334 1335 while ((bio = bio_list_pop(&bio_list))) 1336 bio_put(bio); 1337 } 1338 1339 /* 1340 * helper to find the stripe number for a given bio. Used to figure out which 1341 * stripe has failed. This expects the bio to correspond to a physical disk, 1342 * so it looks up based on physical sector numbers. 1343 */ 1344 static int find_bio_stripe(struct btrfs_raid_bio *rbio, 1345 struct bio *bio) 1346 { 1347 u64 physical = bio->bi_iter.bi_sector; 1348 u64 stripe_start; 1349 int i; 1350 struct btrfs_bio_stripe *stripe; 1351 1352 physical <<= 9; 1353 1354 for (i = 0; i < rbio->bbio->num_stripes; i++) { 1355 stripe = &rbio->bbio->stripes[i]; 1356 stripe_start = stripe->physical; 1357 if (physical >= stripe_start && 1358 physical < stripe_start + rbio->stripe_len && 1359 stripe->dev->bdev && 1360 bio->bi_disk == stripe->dev->bdev->bd_disk && 1361 bio->bi_partno == stripe->dev->bdev->bd_partno) { 1362 return i; 1363 } 1364 } 1365 return -1; 1366 } 1367 1368 /* 1369 * helper to find the stripe number for a given 1370 * bio (before mapping). Used to figure out which stripe has 1371 * failed. This looks up based on logical block numbers. 1372 */ 1373 static int find_logical_bio_stripe(struct btrfs_raid_bio *rbio, 1374 struct bio *bio) 1375 { 1376 u64 logical = bio->bi_iter.bi_sector; 1377 u64 stripe_start; 1378 int i; 1379 1380 logical <<= 9; 1381 1382 for (i = 0; i < rbio->nr_data; i++) { 1383 stripe_start = rbio->bbio->raid_map[i]; 1384 if (logical >= stripe_start && 1385 logical < stripe_start + rbio->stripe_len) { 1386 return i; 1387 } 1388 } 1389 return -1; 1390 } 1391 1392 /* 1393 * returns -EIO if we had too many failures 1394 */ 1395 static int fail_rbio_index(struct btrfs_raid_bio *rbio, int failed) 1396 { 1397 unsigned long flags; 1398 int ret = 0; 1399 1400 spin_lock_irqsave(&rbio->bio_list_lock, flags); 1401 1402 /* we already know this stripe is bad, move on */ 1403 if (rbio->faila == failed || rbio->failb == failed) 1404 goto out; 1405 1406 if (rbio->faila == -1) { 1407 /* first failure on this rbio */ 1408 rbio->faila = failed; 1409 atomic_inc(&rbio->error); 1410 } else if (rbio->failb == -1) { 1411 /* second failure on this rbio */ 1412 rbio->failb = failed; 1413 atomic_inc(&rbio->error); 1414 } else { 1415 ret = -EIO; 1416 } 1417 out: 1418 spin_unlock_irqrestore(&rbio->bio_list_lock, flags); 1419 1420 return ret; 1421 } 1422 1423 /* 1424 * helper to fail a stripe based on a physical disk 1425 * bio. 1426 */ 1427 static int fail_bio_stripe(struct btrfs_raid_bio *rbio, 1428 struct bio *bio) 1429 { 1430 int failed = find_bio_stripe(rbio, bio); 1431 1432 if (failed < 0) 1433 return -EIO; 1434 1435 return fail_rbio_index(rbio, failed); 1436 } 1437 1438 /* 1439 * this sets each page in the bio uptodate. It should only be used on private 1440 * rbio pages, nothing that comes in from the higher layers 1441 */ 1442 static void set_bio_pages_uptodate(struct bio *bio) 1443 { 1444 struct bio_vec *bvec; 1445 int i; 1446 1447 ASSERT(!bio_flagged(bio, BIO_CLONED)); 1448 1449 bio_for_each_segment_all(bvec, bio, i) 1450 SetPageUptodate(bvec->bv_page); 1451 } 1452 1453 /* 1454 * end io for the read phase of the rmw cycle. All the bios here are physical 1455 * stripe bios we've read from the disk so we can recalculate the parity of the 1456 * stripe. 1457 * 1458 * This will usually kick off finish_rmw once all the bios are read in, but it 1459 * may trigger parity reconstruction if we had any errors along the way 1460 */ 1461 static void raid_rmw_end_io(struct bio *bio) 1462 { 1463 struct btrfs_raid_bio *rbio = bio->bi_private; 1464 1465 if (bio->bi_status) 1466 fail_bio_stripe(rbio, bio); 1467 else 1468 set_bio_pages_uptodate(bio); 1469 1470 bio_put(bio); 1471 1472 if (!atomic_dec_and_test(&rbio->stripes_pending)) 1473 return; 1474 1475 if (atomic_read(&rbio->error) > rbio->bbio->max_errors) 1476 goto cleanup; 1477 1478 /* 1479 * this will normally call finish_rmw to start our write 1480 * but if there are any failed stripes we'll reconstruct 1481 * from parity first 1482 */ 1483 validate_rbio_for_rmw(rbio); 1484 return; 1485 1486 cleanup: 1487 1488 rbio_orig_end_io(rbio, BLK_STS_IOERR); 1489 } 1490 1491 /* 1492 * the stripe must be locked by the caller. It will 1493 * unlock after all the writes are done 1494 */ 1495 static int raid56_rmw_stripe(struct btrfs_raid_bio *rbio) 1496 { 1497 int bios_to_read = 0; 1498 struct bio_list bio_list; 1499 int ret; 1500 int pagenr; 1501 int stripe; 1502 struct bio *bio; 1503 1504 bio_list_init(&bio_list); 1505 1506 ret = alloc_rbio_pages(rbio); 1507 if (ret) 1508 goto cleanup; 1509 1510 index_rbio_pages(rbio); 1511 1512 atomic_set(&rbio->error, 0); 1513 /* 1514 * build a list of bios to read all the missing parts of this 1515 * stripe 1516 */ 1517 for (stripe = 0; stripe < rbio->nr_data; stripe++) { 1518 for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { 1519 struct page *page; 1520 /* 1521 * we want to find all the pages missing from 1522 * the rbio and read them from the disk. If 1523 * page_in_rbio finds a page in the bio list 1524 * we don't need to read it off the stripe. 1525 */ 1526 page = page_in_rbio(rbio, stripe, pagenr, 1); 1527 if (page) 1528 continue; 1529 1530 page = rbio_stripe_page(rbio, stripe, pagenr); 1531 /* 1532 * the bio cache may have handed us an uptodate 1533 * page. If so, be happy and use it 1534 */ 1535 if (PageUptodate(page)) 1536 continue; 1537 1538 ret = rbio_add_io_page(rbio, &bio_list, page, 1539 stripe, pagenr, rbio->stripe_len); 1540 if (ret) 1541 goto cleanup; 1542 } 1543 } 1544 1545 bios_to_read = bio_list_size(&bio_list); 1546 if (!bios_to_read) { 1547 /* 1548 * this can happen if others have merged with 1549 * us, it means there is nothing left to read. 1550 * But if there are missing devices it may not be 1551 * safe to do the full stripe write yet. 1552 */ 1553 goto finish; 1554 } 1555 1556 /* 1557 * the bbio may be freed once we submit the last bio. Make sure 1558 * not to touch it after that 1559 */ 1560 atomic_set(&rbio->stripes_pending, bios_to_read); 1561 while (1) { 1562 bio = bio_list_pop(&bio_list); 1563 if (!bio) 1564 break; 1565 1566 bio->bi_private = rbio; 1567 bio->bi_end_io = raid_rmw_end_io; 1568 bio->bi_opf = REQ_OP_READ; 1569 1570 btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); 1571 1572 submit_bio(bio); 1573 } 1574 /* the actual write will happen once the reads are done */ 1575 return 0; 1576 1577 cleanup: 1578 rbio_orig_end_io(rbio, BLK_STS_IOERR); 1579 1580 while ((bio = bio_list_pop(&bio_list))) 1581 bio_put(bio); 1582 1583 return -EIO; 1584 1585 finish: 1586 validate_rbio_for_rmw(rbio); 1587 return 0; 1588 } 1589 1590 /* 1591 * if the upper layers pass in a full stripe, we thank them by only allocating 1592 * enough pages to hold the parity, and sending it all down quickly. 1593 */ 1594 static int full_stripe_write(struct btrfs_raid_bio *rbio) 1595 { 1596 int ret; 1597 1598 ret = alloc_rbio_parity_pages(rbio); 1599 if (ret) { 1600 __free_raid_bio(rbio); 1601 return ret; 1602 } 1603 1604 ret = lock_stripe_add(rbio); 1605 if (ret == 0) 1606 finish_rmw(rbio); 1607 return 0; 1608 } 1609 1610 /* 1611 * partial stripe writes get handed over to async helpers. 1612 * We're really hoping to merge a few more writes into this 1613 * rbio before calculating new parity 1614 */ 1615 static int partial_stripe_write(struct btrfs_raid_bio *rbio) 1616 { 1617 int ret; 1618 1619 ret = lock_stripe_add(rbio); 1620 if (ret == 0) 1621 start_async_work(rbio, rmw_work); 1622 return 0; 1623 } 1624 1625 /* 1626 * sometimes while we were reading from the drive to 1627 * recalculate parity, enough new bios come into create 1628 * a full stripe. So we do a check here to see if we can 1629 * go directly to finish_rmw 1630 */ 1631 static int __raid56_parity_write(struct btrfs_raid_bio *rbio) 1632 { 1633 /* head off into rmw land if we don't have a full stripe */ 1634 if (!rbio_is_full(rbio)) 1635 return partial_stripe_write(rbio); 1636 return full_stripe_write(rbio); 1637 } 1638 1639 /* 1640 * We use plugging call backs to collect full stripes. 1641 * Any time we get a partial stripe write while plugged 1642 * we collect it into a list. When the unplug comes down, 1643 * we sort the list by logical block number and merge 1644 * everything we can into the same rbios 1645 */ 1646 struct btrfs_plug_cb { 1647 struct blk_plug_cb cb; 1648 struct btrfs_fs_info *info; 1649 struct list_head rbio_list; 1650 struct btrfs_work work; 1651 }; 1652 1653 /* 1654 * rbios on the plug list are sorted for easier merging. 1655 */ 1656 static int plug_cmp(void *priv, struct list_head *a, struct list_head *b) 1657 { 1658 struct btrfs_raid_bio *ra = container_of(a, struct btrfs_raid_bio, 1659 plug_list); 1660 struct btrfs_raid_bio *rb = container_of(b, struct btrfs_raid_bio, 1661 plug_list); 1662 u64 a_sector = ra->bio_list.head->bi_iter.bi_sector; 1663 u64 b_sector = rb->bio_list.head->bi_iter.bi_sector; 1664 1665 if (a_sector < b_sector) 1666 return -1; 1667 if (a_sector > b_sector) 1668 return 1; 1669 return 0; 1670 } 1671 1672 static void run_plug(struct btrfs_plug_cb *plug) 1673 { 1674 struct btrfs_raid_bio *cur; 1675 struct btrfs_raid_bio *last = NULL; 1676 1677 /* 1678 * sort our plug list then try to merge 1679 * everything we can in hopes of creating full 1680 * stripes. 1681 */ 1682 list_sort(NULL, &plug->rbio_list, plug_cmp); 1683 while (!list_empty(&plug->rbio_list)) { 1684 cur = list_entry(plug->rbio_list.next, 1685 struct btrfs_raid_bio, plug_list); 1686 list_del_init(&cur->plug_list); 1687 1688 if (rbio_is_full(cur)) { 1689 int ret; 1690 1691 /* we have a full stripe, send it down */ 1692 ret = full_stripe_write(cur); 1693 BUG_ON(ret); 1694 continue; 1695 } 1696 if (last) { 1697 if (rbio_can_merge(last, cur)) { 1698 merge_rbio(last, cur); 1699 __free_raid_bio(cur); 1700 continue; 1701 1702 } 1703 __raid56_parity_write(last); 1704 } 1705 last = cur; 1706 } 1707 if (last) { 1708 __raid56_parity_write(last); 1709 } 1710 kfree(plug); 1711 } 1712 1713 /* 1714 * if the unplug comes from schedule, we have to push the 1715 * work off to a helper thread 1716 */ 1717 static void unplug_work(struct btrfs_work *work) 1718 { 1719 struct btrfs_plug_cb *plug; 1720 plug = container_of(work, struct btrfs_plug_cb, work); 1721 run_plug(plug); 1722 } 1723 1724 static void btrfs_raid_unplug(struct blk_plug_cb *cb, bool from_schedule) 1725 { 1726 struct btrfs_plug_cb *plug; 1727 plug = container_of(cb, struct btrfs_plug_cb, cb); 1728 1729 if (from_schedule) { 1730 btrfs_init_work(&plug->work, btrfs_rmw_helper, 1731 unplug_work, NULL, NULL); 1732 btrfs_queue_work(plug->info->rmw_workers, 1733 &plug->work); 1734 return; 1735 } 1736 run_plug(plug); 1737 } 1738 1739 /* 1740 * our main entry point for writes from the rest of the FS. 1741 */ 1742 int raid56_parity_write(struct btrfs_fs_info *fs_info, struct bio *bio, 1743 struct btrfs_bio *bbio, u64 stripe_len) 1744 { 1745 struct btrfs_raid_bio *rbio; 1746 struct btrfs_plug_cb *plug = NULL; 1747 struct blk_plug_cb *cb; 1748 int ret; 1749 1750 rbio = alloc_rbio(fs_info, bbio, stripe_len); 1751 if (IS_ERR(rbio)) { 1752 btrfs_put_bbio(bbio); 1753 return PTR_ERR(rbio); 1754 } 1755 bio_list_add(&rbio->bio_list, bio); 1756 rbio->bio_list_bytes = bio->bi_iter.bi_size; 1757 rbio->operation = BTRFS_RBIO_WRITE; 1758 1759 btrfs_bio_counter_inc_noblocked(fs_info); 1760 rbio->generic_bio_cnt = 1; 1761 1762 /* 1763 * don't plug on full rbios, just get them out the door 1764 * as quickly as we can 1765 */ 1766 if (rbio_is_full(rbio)) { 1767 ret = full_stripe_write(rbio); 1768 if (ret) 1769 btrfs_bio_counter_dec(fs_info); 1770 return ret; 1771 } 1772 1773 cb = blk_check_plugged(btrfs_raid_unplug, fs_info, sizeof(*plug)); 1774 if (cb) { 1775 plug = container_of(cb, struct btrfs_plug_cb, cb); 1776 if (!plug->info) { 1777 plug->info = fs_info; 1778 INIT_LIST_HEAD(&plug->rbio_list); 1779 } 1780 list_add_tail(&rbio->plug_list, &plug->rbio_list); 1781 ret = 0; 1782 } else { 1783 ret = __raid56_parity_write(rbio); 1784 if (ret) 1785 btrfs_bio_counter_dec(fs_info); 1786 } 1787 return ret; 1788 } 1789 1790 /* 1791 * all parity reconstruction happens here. We've read in everything 1792 * we can find from the drives and this does the heavy lifting of 1793 * sorting the good from the bad. 1794 */ 1795 static void __raid_recover_end_io(struct btrfs_raid_bio *rbio) 1796 { 1797 int pagenr, stripe; 1798 void **pointers; 1799 int faila = -1, failb = -1; 1800 struct page *page; 1801 blk_status_t err; 1802 int i; 1803 1804 pointers = kcalloc(rbio->real_stripes, sizeof(void *), GFP_NOFS); 1805 if (!pointers) { 1806 err = BLK_STS_RESOURCE; 1807 goto cleanup_io; 1808 } 1809 1810 faila = rbio->faila; 1811 failb = rbio->failb; 1812 1813 if (rbio->operation == BTRFS_RBIO_READ_REBUILD || 1814 rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { 1815 spin_lock_irq(&rbio->bio_list_lock); 1816 set_bit(RBIO_RMW_LOCKED_BIT, &rbio->flags); 1817 spin_unlock_irq(&rbio->bio_list_lock); 1818 } 1819 1820 index_rbio_pages(rbio); 1821 1822 for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { 1823 /* 1824 * Now we just use bitmap to mark the horizontal stripes in 1825 * which we have data when doing parity scrub. 1826 */ 1827 if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB && 1828 !test_bit(pagenr, rbio->dbitmap)) 1829 continue; 1830 1831 /* setup our array of pointers with pages 1832 * from each stripe 1833 */ 1834 for (stripe = 0; stripe < rbio->real_stripes; stripe++) { 1835 /* 1836 * if we're rebuilding a read, we have to use 1837 * pages from the bio list 1838 */ 1839 if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || 1840 rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && 1841 (stripe == faila || stripe == failb)) { 1842 page = page_in_rbio(rbio, stripe, pagenr, 0); 1843 } else { 1844 page = rbio_stripe_page(rbio, stripe, pagenr); 1845 } 1846 pointers[stripe] = kmap(page); 1847 } 1848 1849 /* all raid6 handling here */ 1850 if (rbio->bbio->map_type & BTRFS_BLOCK_GROUP_RAID6) { 1851 /* 1852 * single failure, rebuild from parity raid5 1853 * style 1854 */ 1855 if (failb < 0) { 1856 if (faila == rbio->nr_data) { 1857 /* 1858 * Just the P stripe has failed, without 1859 * a bad data or Q stripe. 1860 * TODO, we should redo the xor here. 1861 */ 1862 err = BLK_STS_IOERR; 1863 goto cleanup; 1864 } 1865 /* 1866 * a single failure in raid6 is rebuilt 1867 * in the pstripe code below 1868 */ 1869 goto pstripe; 1870 } 1871 1872 /* make sure our ps and qs are in order */ 1873 if (faila > failb) { 1874 int tmp = failb; 1875 failb = faila; 1876 faila = tmp; 1877 } 1878 1879 /* if the q stripe is failed, do a pstripe reconstruction 1880 * from the xors. 1881 * If both the q stripe and the P stripe are failed, we're 1882 * here due to a crc mismatch and we can't give them the 1883 * data they want 1884 */ 1885 if (rbio->bbio->raid_map[failb] == RAID6_Q_STRIPE) { 1886 if (rbio->bbio->raid_map[faila] == 1887 RAID5_P_STRIPE) { 1888 err = BLK_STS_IOERR; 1889 goto cleanup; 1890 } 1891 /* 1892 * otherwise we have one bad data stripe and 1893 * a good P stripe. raid5! 1894 */ 1895 goto pstripe; 1896 } 1897 1898 if (rbio->bbio->raid_map[failb] == RAID5_P_STRIPE) { 1899 raid6_datap_recov(rbio->real_stripes, 1900 PAGE_SIZE, faila, pointers); 1901 } else { 1902 raid6_2data_recov(rbio->real_stripes, 1903 PAGE_SIZE, faila, failb, 1904 pointers); 1905 } 1906 } else { 1907 void *p; 1908 1909 /* rebuild from P stripe here (raid5 or raid6) */ 1910 BUG_ON(failb != -1); 1911 pstripe: 1912 /* Copy parity block into failed block to start with */ 1913 copy_page(pointers[faila], pointers[rbio->nr_data]); 1914 1915 /* rearrange the pointer array */ 1916 p = pointers[faila]; 1917 for (stripe = faila; stripe < rbio->nr_data - 1; stripe++) 1918 pointers[stripe] = pointers[stripe + 1]; 1919 pointers[rbio->nr_data - 1] = p; 1920 1921 /* xor in the rest */ 1922 run_xor(pointers, rbio->nr_data - 1, PAGE_SIZE); 1923 } 1924 /* if we're doing this rebuild as part of an rmw, go through 1925 * and set all of our private rbio pages in the 1926 * failed stripes as uptodate. This way finish_rmw will 1927 * know they can be trusted. If this was a read reconstruction, 1928 * other endio functions will fiddle the uptodate bits 1929 */ 1930 if (rbio->operation == BTRFS_RBIO_WRITE) { 1931 for (i = 0; i < rbio->stripe_npages; i++) { 1932 if (faila != -1) { 1933 page = rbio_stripe_page(rbio, faila, i); 1934 SetPageUptodate(page); 1935 } 1936 if (failb != -1) { 1937 page = rbio_stripe_page(rbio, failb, i); 1938 SetPageUptodate(page); 1939 } 1940 } 1941 } 1942 for (stripe = 0; stripe < rbio->real_stripes; stripe++) { 1943 /* 1944 * if we're rebuilding a read, we have to use 1945 * pages from the bio list 1946 */ 1947 if ((rbio->operation == BTRFS_RBIO_READ_REBUILD || 1948 rbio->operation == BTRFS_RBIO_REBUILD_MISSING) && 1949 (stripe == faila || stripe == failb)) { 1950 page = page_in_rbio(rbio, stripe, pagenr, 0); 1951 } else { 1952 page = rbio_stripe_page(rbio, stripe, pagenr); 1953 } 1954 kunmap(page); 1955 } 1956 } 1957 1958 err = BLK_STS_OK; 1959 cleanup: 1960 kfree(pointers); 1961 1962 cleanup_io: 1963 /* 1964 * Similar to READ_REBUILD, REBUILD_MISSING at this point also has a 1965 * valid rbio which is consistent with ondisk content, thus such a 1966 * valid rbio can be cached to avoid further disk reads. 1967 */ 1968 if (rbio->operation == BTRFS_RBIO_READ_REBUILD || 1969 rbio->operation == BTRFS_RBIO_REBUILD_MISSING) { 1970 /* 1971 * - In case of two failures, where rbio->failb != -1: 1972 * 1973 * Do not cache this rbio since the above read reconstruction 1974 * (raid6_datap_recov() or raid6_2data_recov()) may have 1975 * changed some content of stripes which are not identical to 1976 * on-disk content any more, otherwise, a later write/recover 1977 * may steal stripe_pages from this rbio and end up with 1978 * corruptions or rebuild failures. 1979 * 1980 * - In case of single failure, where rbio->failb == -1: 1981 * 1982 * Cache this rbio iff the above read reconstruction is 1983 * excuted without problems. 1984 */ 1985 if (err == BLK_STS_OK && rbio->failb < 0) 1986 cache_rbio_pages(rbio); 1987 else 1988 clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); 1989 1990 rbio_orig_end_io(rbio, err); 1991 } else if (err == BLK_STS_OK) { 1992 rbio->faila = -1; 1993 rbio->failb = -1; 1994 1995 if (rbio->operation == BTRFS_RBIO_WRITE) 1996 finish_rmw(rbio); 1997 else if (rbio->operation == BTRFS_RBIO_PARITY_SCRUB) 1998 finish_parity_scrub(rbio, 0); 1999 else 2000 BUG(); 2001 } else { 2002 rbio_orig_end_io(rbio, err); 2003 } 2004 } 2005 2006 /* 2007 * This is called only for stripes we've read from disk to 2008 * reconstruct the parity. 2009 */ 2010 static void raid_recover_end_io(struct bio *bio) 2011 { 2012 struct btrfs_raid_bio *rbio = bio->bi_private; 2013 2014 /* 2015 * we only read stripe pages off the disk, set them 2016 * up to date if there were no errors 2017 */ 2018 if (bio->bi_status) 2019 fail_bio_stripe(rbio, bio); 2020 else 2021 set_bio_pages_uptodate(bio); 2022 bio_put(bio); 2023 2024 if (!atomic_dec_and_test(&rbio->stripes_pending)) 2025 return; 2026 2027 if (atomic_read(&rbio->error) > rbio->bbio->max_errors) 2028 rbio_orig_end_io(rbio, BLK_STS_IOERR); 2029 else 2030 __raid_recover_end_io(rbio); 2031 } 2032 2033 /* 2034 * reads everything we need off the disk to reconstruct 2035 * the parity. endio handlers trigger final reconstruction 2036 * when the IO is done. 2037 * 2038 * This is used both for reads from the higher layers and for 2039 * parity construction required to finish a rmw cycle. 2040 */ 2041 static int __raid56_parity_recover(struct btrfs_raid_bio *rbio) 2042 { 2043 int bios_to_read = 0; 2044 struct bio_list bio_list; 2045 int ret; 2046 int pagenr; 2047 int stripe; 2048 struct bio *bio; 2049 2050 bio_list_init(&bio_list); 2051 2052 ret = alloc_rbio_pages(rbio); 2053 if (ret) 2054 goto cleanup; 2055 2056 atomic_set(&rbio->error, 0); 2057 2058 /* 2059 * read everything that hasn't failed. Thanks to the 2060 * stripe cache, it is possible that some or all of these 2061 * pages are going to be uptodate. 2062 */ 2063 for (stripe = 0; stripe < rbio->real_stripes; stripe++) { 2064 if (rbio->faila == stripe || rbio->failb == stripe) { 2065 atomic_inc(&rbio->error); 2066 continue; 2067 } 2068 2069 for (pagenr = 0; pagenr < rbio->stripe_npages; pagenr++) { 2070 struct page *p; 2071 2072 /* 2073 * the rmw code may have already read this 2074 * page in 2075 */ 2076 p = rbio_stripe_page(rbio, stripe, pagenr); 2077 if (PageUptodate(p)) 2078 continue; 2079 2080 ret = rbio_add_io_page(rbio, &bio_list, 2081 rbio_stripe_page(rbio, stripe, pagenr), 2082 stripe, pagenr, rbio->stripe_len); 2083 if (ret < 0) 2084 goto cleanup; 2085 } 2086 } 2087 2088 bios_to_read = bio_list_size(&bio_list); 2089 if (!bios_to_read) { 2090 /* 2091 * we might have no bios to read just because the pages 2092 * were up to date, or we might have no bios to read because 2093 * the devices were gone. 2094 */ 2095 if (atomic_read(&rbio->error) <= rbio->bbio->max_errors) { 2096 __raid_recover_end_io(rbio); 2097 goto out; 2098 } else { 2099 goto cleanup; 2100 } 2101 } 2102 2103 /* 2104 * the bbio may be freed once we submit the last bio. Make sure 2105 * not to touch it after that 2106 */ 2107 atomic_set(&rbio->stripes_pending, bios_to_read); 2108 while (1) { 2109 bio = bio_list_pop(&bio_list); 2110 if (!bio) 2111 break; 2112 2113 bio->bi_private = rbio; 2114 bio->bi_end_io = raid_recover_end_io; 2115 bio->bi_opf = REQ_OP_READ; 2116 2117 btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); 2118 2119 submit_bio(bio); 2120 } 2121 out: 2122 return 0; 2123 2124 cleanup: 2125 if (rbio->operation == BTRFS_RBIO_READ_REBUILD || 2126 rbio->operation == BTRFS_RBIO_REBUILD_MISSING) 2127 rbio_orig_end_io(rbio, BLK_STS_IOERR); 2128 2129 while ((bio = bio_list_pop(&bio_list))) 2130 bio_put(bio); 2131 2132 return -EIO; 2133 } 2134 2135 /* 2136 * the main entry point for reads from the higher layers. This 2137 * is really only called when the normal read path had a failure, 2138 * so we assume the bio they send down corresponds to a failed part 2139 * of the drive. 2140 */ 2141 int raid56_parity_recover(struct btrfs_fs_info *fs_info, struct bio *bio, 2142 struct btrfs_bio *bbio, u64 stripe_len, 2143 int mirror_num, int generic_io) 2144 { 2145 struct btrfs_raid_bio *rbio; 2146 int ret; 2147 2148 if (generic_io) { 2149 ASSERT(bbio->mirror_num == mirror_num); 2150 btrfs_io_bio(bio)->mirror_num = mirror_num; 2151 } 2152 2153 rbio = alloc_rbio(fs_info, bbio, stripe_len); 2154 if (IS_ERR(rbio)) { 2155 if (generic_io) 2156 btrfs_put_bbio(bbio); 2157 return PTR_ERR(rbio); 2158 } 2159 2160 rbio->operation = BTRFS_RBIO_READ_REBUILD; 2161 bio_list_add(&rbio->bio_list, bio); 2162 rbio->bio_list_bytes = bio->bi_iter.bi_size; 2163 2164 rbio->faila = find_logical_bio_stripe(rbio, bio); 2165 if (rbio->faila == -1) { 2166 btrfs_warn(fs_info, 2167 "%s could not find the bad stripe in raid56 so that we cannot recover any more (bio has logical %llu len %llu, bbio has map_type %llu)", 2168 __func__, (u64)bio->bi_iter.bi_sector << 9, 2169 (u64)bio->bi_iter.bi_size, bbio->map_type); 2170 if (generic_io) 2171 btrfs_put_bbio(bbio); 2172 kfree(rbio); 2173 return -EIO; 2174 } 2175 2176 if (generic_io) { 2177 btrfs_bio_counter_inc_noblocked(fs_info); 2178 rbio->generic_bio_cnt = 1; 2179 } else { 2180 btrfs_get_bbio(bbio); 2181 } 2182 2183 /* 2184 * Loop retry: 2185 * for 'mirror == 2', reconstruct from all other stripes. 2186 * for 'mirror_num > 2', select a stripe to fail on every retry. 2187 */ 2188 if (mirror_num > 2) { 2189 /* 2190 * 'mirror == 3' is to fail the p stripe and 2191 * reconstruct from the q stripe. 'mirror > 3' is to 2192 * fail a data stripe and reconstruct from p+q stripe. 2193 */ 2194 rbio->failb = rbio->real_stripes - (mirror_num - 1); 2195 ASSERT(rbio->failb > 0); 2196 if (rbio->failb <= rbio->faila) 2197 rbio->failb--; 2198 } 2199 2200 ret = lock_stripe_add(rbio); 2201 2202 /* 2203 * __raid56_parity_recover will end the bio with 2204 * any errors it hits. We don't want to return 2205 * its error value up the stack because our caller 2206 * will end up calling bio_endio with any nonzero 2207 * return 2208 */ 2209 if (ret == 0) 2210 __raid56_parity_recover(rbio); 2211 /* 2212 * our rbio has been added to the list of 2213 * rbios that will be handled after the 2214 * currently lock owner is done 2215 */ 2216 return 0; 2217 2218 } 2219 2220 static void rmw_work(struct btrfs_work *work) 2221 { 2222 struct btrfs_raid_bio *rbio; 2223 2224 rbio = container_of(work, struct btrfs_raid_bio, work); 2225 raid56_rmw_stripe(rbio); 2226 } 2227 2228 static void read_rebuild_work(struct btrfs_work *work) 2229 { 2230 struct btrfs_raid_bio *rbio; 2231 2232 rbio = container_of(work, struct btrfs_raid_bio, work); 2233 __raid56_parity_recover(rbio); 2234 } 2235 2236 /* 2237 * The following code is used to scrub/replace the parity stripe 2238 * 2239 * Caller must have already increased bio_counter for getting @bbio. 2240 * 2241 * Note: We need make sure all the pages that add into the scrub/replace 2242 * raid bio are correct and not be changed during the scrub/replace. That 2243 * is those pages just hold metadata or file data with checksum. 2244 */ 2245 2246 struct btrfs_raid_bio * 2247 raid56_parity_alloc_scrub_rbio(struct btrfs_fs_info *fs_info, struct bio *bio, 2248 struct btrfs_bio *bbio, u64 stripe_len, 2249 struct btrfs_device *scrub_dev, 2250 unsigned long *dbitmap, int stripe_nsectors) 2251 { 2252 struct btrfs_raid_bio *rbio; 2253 int i; 2254 2255 rbio = alloc_rbio(fs_info, bbio, stripe_len); 2256 if (IS_ERR(rbio)) 2257 return NULL; 2258 bio_list_add(&rbio->bio_list, bio); 2259 /* 2260 * This is a special bio which is used to hold the completion handler 2261 * and make the scrub rbio is similar to the other types 2262 */ 2263 ASSERT(!bio->bi_iter.bi_size); 2264 rbio->operation = BTRFS_RBIO_PARITY_SCRUB; 2265 2266 /* 2267 * After mapping bbio with BTRFS_MAP_WRITE, parities have been sorted 2268 * to the end position, so this search can start from the first parity 2269 * stripe. 2270 */ 2271 for (i = rbio->nr_data; i < rbio->real_stripes; i++) { 2272 if (bbio->stripes[i].dev == scrub_dev) { 2273 rbio->scrubp = i; 2274 break; 2275 } 2276 } 2277 ASSERT(i < rbio->real_stripes); 2278 2279 /* Now we just support the sectorsize equals to page size */ 2280 ASSERT(fs_info->sectorsize == PAGE_SIZE); 2281 ASSERT(rbio->stripe_npages == stripe_nsectors); 2282 bitmap_copy(rbio->dbitmap, dbitmap, stripe_nsectors); 2283 2284 /* 2285 * We have already increased bio_counter when getting bbio, record it 2286 * so we can free it at rbio_orig_end_io(). 2287 */ 2288 rbio->generic_bio_cnt = 1; 2289 2290 return rbio; 2291 } 2292 2293 /* Used for both parity scrub and missing. */ 2294 void raid56_add_scrub_pages(struct btrfs_raid_bio *rbio, struct page *page, 2295 u64 logical) 2296 { 2297 int stripe_offset; 2298 int index; 2299 2300 ASSERT(logical >= rbio->bbio->raid_map[0]); 2301 ASSERT(logical + PAGE_SIZE <= rbio->bbio->raid_map[0] + 2302 rbio->stripe_len * rbio->nr_data); 2303 stripe_offset = (int)(logical - rbio->bbio->raid_map[0]); 2304 index = stripe_offset >> PAGE_SHIFT; 2305 rbio->bio_pages[index] = page; 2306 } 2307 2308 /* 2309 * We just scrub the parity that we have correct data on the same horizontal, 2310 * so we needn't allocate all pages for all the stripes. 2311 */ 2312 static int alloc_rbio_essential_pages(struct btrfs_raid_bio *rbio) 2313 { 2314 int i; 2315 int bit; 2316 int index; 2317 struct page *page; 2318 2319 for_each_set_bit(bit, rbio->dbitmap, rbio->stripe_npages) { 2320 for (i = 0; i < rbio->real_stripes; i++) { 2321 index = i * rbio->stripe_npages + bit; 2322 if (rbio->stripe_pages[index]) 2323 continue; 2324 2325 page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 2326 if (!page) 2327 return -ENOMEM; 2328 rbio->stripe_pages[index] = page; 2329 } 2330 } 2331 return 0; 2332 } 2333 2334 static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio, 2335 int need_check) 2336 { 2337 struct btrfs_bio *bbio = rbio->bbio; 2338 void **pointers = rbio->finish_pointers; 2339 unsigned long *pbitmap = rbio->finish_pbitmap; 2340 int nr_data = rbio->nr_data; 2341 int stripe; 2342 int pagenr; 2343 int p_stripe = -1; 2344 int q_stripe = -1; 2345 struct page *p_page = NULL; 2346 struct page *q_page = NULL; 2347 struct bio_list bio_list; 2348 struct bio *bio; 2349 int is_replace = 0; 2350 int ret; 2351 2352 bio_list_init(&bio_list); 2353 2354 if (rbio->real_stripes - rbio->nr_data == 1) { 2355 p_stripe = rbio->real_stripes - 1; 2356 } else if (rbio->real_stripes - rbio->nr_data == 2) { 2357 p_stripe = rbio->real_stripes - 2; 2358 q_stripe = rbio->real_stripes - 1; 2359 } else { 2360 BUG(); 2361 } 2362 2363 if (bbio->num_tgtdevs && bbio->tgtdev_map[rbio->scrubp]) { 2364 is_replace = 1; 2365 bitmap_copy(pbitmap, rbio->dbitmap, rbio->stripe_npages); 2366 } 2367 2368 /* 2369 * Because the higher layers(scrubber) are unlikely to 2370 * use this area of the disk again soon, so don't cache 2371 * it. 2372 */ 2373 clear_bit(RBIO_CACHE_READY_BIT, &rbio->flags); 2374 2375 if (!need_check) 2376 goto writeback; 2377 2378 p_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 2379 if (!p_page) 2380 goto cleanup; 2381 SetPageUptodate(p_page); 2382 2383 if (q_stripe != -1) { 2384 q_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); 2385 if (!q_page) { 2386 __free_page(p_page); 2387 goto cleanup; 2388 } 2389 SetPageUptodate(q_page); 2390 } 2391 2392 atomic_set(&rbio->error, 0); 2393 2394 for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { 2395 struct page *p; 2396 void *parity; 2397 /* first collect one page from each data stripe */ 2398 for (stripe = 0; stripe < nr_data; stripe++) { 2399 p = page_in_rbio(rbio, stripe, pagenr, 0); 2400 pointers[stripe] = kmap(p); 2401 } 2402 2403 /* then add the parity stripe */ 2404 pointers[stripe++] = kmap(p_page); 2405 2406 if (q_stripe != -1) { 2407 2408 /* 2409 * raid6, add the qstripe and call the 2410 * library function to fill in our p/q 2411 */ 2412 pointers[stripe++] = kmap(q_page); 2413 2414 raid6_call.gen_syndrome(rbio->real_stripes, PAGE_SIZE, 2415 pointers); 2416 } else { 2417 /* raid5 */ 2418 copy_page(pointers[nr_data], pointers[0]); 2419 run_xor(pointers + 1, nr_data - 1, PAGE_SIZE); 2420 } 2421 2422 /* Check scrubbing parity and repair it */ 2423 p = rbio_stripe_page(rbio, rbio->scrubp, pagenr); 2424 parity = kmap(p); 2425 if (memcmp(parity, pointers[rbio->scrubp], PAGE_SIZE)) 2426 copy_page(parity, pointers[rbio->scrubp]); 2427 else 2428 /* Parity is right, needn't writeback */ 2429 bitmap_clear(rbio->dbitmap, pagenr, 1); 2430 kunmap(p); 2431 2432 for (stripe = 0; stripe < rbio->real_stripes; stripe++) 2433 kunmap(page_in_rbio(rbio, stripe, pagenr, 0)); 2434 } 2435 2436 __free_page(p_page); 2437 if (q_page) 2438 __free_page(q_page); 2439 2440 writeback: 2441 /* 2442 * time to start writing. Make bios for everything from the 2443 * higher layers (the bio_list in our rbio) and our p/q. Ignore 2444 * everything else. 2445 */ 2446 for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { 2447 struct page *page; 2448 2449 page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); 2450 ret = rbio_add_io_page(rbio, &bio_list, 2451 page, rbio->scrubp, pagenr, rbio->stripe_len); 2452 if (ret) 2453 goto cleanup; 2454 } 2455 2456 if (!is_replace) 2457 goto submit_write; 2458 2459 for_each_set_bit(pagenr, pbitmap, rbio->stripe_npages) { 2460 struct page *page; 2461 2462 page = rbio_stripe_page(rbio, rbio->scrubp, pagenr); 2463 ret = rbio_add_io_page(rbio, &bio_list, page, 2464 bbio->tgtdev_map[rbio->scrubp], 2465 pagenr, rbio->stripe_len); 2466 if (ret) 2467 goto cleanup; 2468 } 2469 2470 submit_write: 2471 nr_data = bio_list_size(&bio_list); 2472 if (!nr_data) { 2473 /* Every parity is right */ 2474 rbio_orig_end_io(rbio, BLK_STS_OK); 2475 return; 2476 } 2477 2478 atomic_set(&rbio->stripes_pending, nr_data); 2479 2480 while (1) { 2481 bio = bio_list_pop(&bio_list); 2482 if (!bio) 2483 break; 2484 2485 bio->bi_private = rbio; 2486 bio->bi_end_io = raid_write_end_io; 2487 bio->bi_opf = REQ_OP_WRITE; 2488 2489 submit_bio(bio); 2490 } 2491 return; 2492 2493 cleanup: 2494 rbio_orig_end_io(rbio, BLK_STS_IOERR); 2495 2496 while ((bio = bio_list_pop(&bio_list))) 2497 bio_put(bio); 2498 } 2499 2500 static inline int is_data_stripe(struct btrfs_raid_bio *rbio, int stripe) 2501 { 2502 if (stripe >= 0 && stripe < rbio->nr_data) 2503 return 1; 2504 return 0; 2505 } 2506 2507 /* 2508 * While we're doing the parity check and repair, we could have errors 2509 * in reading pages off the disk. This checks for errors and if we're 2510 * not able to read the page it'll trigger parity reconstruction. The 2511 * parity scrub will be finished after we've reconstructed the failed 2512 * stripes 2513 */ 2514 static void validate_rbio_for_parity_scrub(struct btrfs_raid_bio *rbio) 2515 { 2516 if (atomic_read(&rbio->error) > rbio->bbio->max_errors) 2517 goto cleanup; 2518 2519 if (rbio->faila >= 0 || rbio->failb >= 0) { 2520 int dfail = 0, failp = -1; 2521 2522 if (is_data_stripe(rbio, rbio->faila)) 2523 dfail++; 2524 else if (is_parity_stripe(rbio->faila)) 2525 failp = rbio->faila; 2526 2527 if (is_data_stripe(rbio, rbio->failb)) 2528 dfail++; 2529 else if (is_parity_stripe(rbio->failb)) 2530 failp = rbio->failb; 2531 2532 /* 2533 * Because we can not use a scrubbing parity to repair 2534 * the data, so the capability of the repair is declined. 2535 * (In the case of RAID5, we can not repair anything) 2536 */ 2537 if (dfail > rbio->bbio->max_errors - 1) 2538 goto cleanup; 2539 2540 /* 2541 * If all data is good, only parity is correctly, just 2542 * repair the parity. 2543 */ 2544 if (dfail == 0) { 2545 finish_parity_scrub(rbio, 0); 2546 return; 2547 } 2548 2549 /* 2550 * Here means we got one corrupted data stripe and one 2551 * corrupted parity on RAID6, if the corrupted parity 2552 * is scrubbing parity, luckily, use the other one to repair 2553 * the data, or we can not repair the data stripe. 2554 */ 2555 if (failp != rbio->scrubp) 2556 goto cleanup; 2557 2558 __raid_recover_end_io(rbio); 2559 } else { 2560 finish_parity_scrub(rbio, 1); 2561 } 2562 return; 2563 2564 cleanup: 2565 rbio_orig_end_io(rbio, BLK_STS_IOERR); 2566 } 2567 2568 /* 2569 * end io for the read phase of the rmw cycle. All the bios here are physical 2570 * stripe bios we've read from the disk so we can recalculate the parity of the 2571 * stripe. 2572 * 2573 * This will usually kick off finish_rmw once all the bios are read in, but it 2574 * may trigger parity reconstruction if we had any errors along the way 2575 */ 2576 static void raid56_parity_scrub_end_io(struct bio *bio) 2577 { 2578 struct btrfs_raid_bio *rbio = bio->bi_private; 2579 2580 if (bio->bi_status) 2581 fail_bio_stripe(rbio, bio); 2582 else 2583 set_bio_pages_uptodate(bio); 2584 2585 bio_put(bio); 2586 2587 if (!atomic_dec_and_test(&rbio->stripes_pending)) 2588 return; 2589 2590 /* 2591 * this will normally call finish_rmw to start our write 2592 * but if there are any failed stripes we'll reconstruct 2593 * from parity first 2594 */ 2595 validate_rbio_for_parity_scrub(rbio); 2596 } 2597 2598 static void raid56_parity_scrub_stripe(struct btrfs_raid_bio *rbio) 2599 { 2600 int bios_to_read = 0; 2601 struct bio_list bio_list; 2602 int ret; 2603 int pagenr; 2604 int stripe; 2605 struct bio *bio; 2606 2607 bio_list_init(&bio_list); 2608 2609 ret = alloc_rbio_essential_pages(rbio); 2610 if (ret) 2611 goto cleanup; 2612 2613 atomic_set(&rbio->error, 0); 2614 /* 2615 * build a list of bios to read all the missing parts of this 2616 * stripe 2617 */ 2618 for (stripe = 0; stripe < rbio->real_stripes; stripe++) { 2619 for_each_set_bit(pagenr, rbio->dbitmap, rbio->stripe_npages) { 2620 struct page *page; 2621 /* 2622 * we want to find all the pages missing from 2623 * the rbio and read them from the disk. If 2624 * page_in_rbio finds a page in the bio list 2625 * we don't need to read it off the stripe. 2626 */ 2627 page = page_in_rbio(rbio, stripe, pagenr, 1); 2628 if (page) 2629 continue; 2630 2631 page = rbio_stripe_page(rbio, stripe, pagenr); 2632 /* 2633 * the bio cache may have handed us an uptodate 2634 * page. If so, be happy and use it 2635 */ 2636 if (PageUptodate(page)) 2637 continue; 2638 2639 ret = rbio_add_io_page(rbio, &bio_list, page, 2640 stripe, pagenr, rbio->stripe_len); 2641 if (ret) 2642 goto cleanup; 2643 } 2644 } 2645 2646 bios_to_read = bio_list_size(&bio_list); 2647 if (!bios_to_read) { 2648 /* 2649 * this can happen if others have merged with 2650 * us, it means there is nothing left to read. 2651 * But if there are missing devices it may not be 2652 * safe to do the full stripe write yet. 2653 */ 2654 goto finish; 2655 } 2656 2657 /* 2658 * the bbio may be freed once we submit the last bio. Make sure 2659 * not to touch it after that 2660 */ 2661 atomic_set(&rbio->stripes_pending, bios_to_read); 2662 while (1) { 2663 bio = bio_list_pop(&bio_list); 2664 if (!bio) 2665 break; 2666 2667 bio->bi_private = rbio; 2668 bio->bi_end_io = raid56_parity_scrub_end_io; 2669 bio->bi_opf = REQ_OP_READ; 2670 2671 btrfs_bio_wq_end_io(rbio->fs_info, bio, BTRFS_WQ_ENDIO_RAID56); 2672 2673 submit_bio(bio); 2674 } 2675 /* the actual write will happen once the reads are done */ 2676 return; 2677 2678 cleanup: 2679 rbio_orig_end_io(rbio, BLK_STS_IOERR); 2680 2681 while ((bio = bio_list_pop(&bio_list))) 2682 bio_put(bio); 2683 2684 return; 2685 2686 finish: 2687 validate_rbio_for_parity_scrub(rbio); 2688 } 2689 2690 static void scrub_parity_work(struct btrfs_work *work) 2691 { 2692 struct btrfs_raid_bio *rbio; 2693 2694 rbio = container_of(work, struct btrfs_raid_bio, work); 2695 raid56_parity_scrub_stripe(rbio); 2696 } 2697 2698 void raid56_parity_submit_scrub_rbio(struct btrfs_raid_bio *rbio) 2699 { 2700 if (!lock_stripe_add(rbio)) 2701 start_async_work(rbio, scrub_parity_work); 2702 } 2703 2704 /* The following code is used for dev replace of a missing RAID 5/6 device. */ 2705 2706 struct btrfs_raid_bio * 2707 raid56_alloc_missing_rbio(struct btrfs_fs_info *fs_info, struct bio *bio, 2708 struct btrfs_bio *bbio, u64 length) 2709 { 2710 struct btrfs_raid_bio *rbio; 2711 2712 rbio = alloc_rbio(fs_info, bbio, length); 2713 if (IS_ERR(rbio)) 2714 return NULL; 2715 2716 rbio->operation = BTRFS_RBIO_REBUILD_MISSING; 2717 bio_list_add(&rbio->bio_list, bio); 2718 /* 2719 * This is a special bio which is used to hold the completion handler 2720 * and make the scrub rbio is similar to the other types 2721 */ 2722 ASSERT(!bio->bi_iter.bi_size); 2723 2724 rbio->faila = find_logical_bio_stripe(rbio, bio); 2725 if (rbio->faila == -1) { 2726 BUG(); 2727 kfree(rbio); 2728 return NULL; 2729 } 2730 2731 /* 2732 * When we get bbio, we have already increased bio_counter, record it 2733 * so we can free it at rbio_orig_end_io() 2734 */ 2735 rbio->generic_bio_cnt = 1; 2736 2737 return rbio; 2738 } 2739 2740 void raid56_submit_missing_rbio(struct btrfs_raid_bio *rbio) 2741 { 2742 if (!lock_stripe_add(rbio)) 2743 start_async_work(rbio, read_rebuild_work); 2744 } 2745